Introduction

The plugin ecosystem of DeepSeek Harness (DSH) allows users to extend functionality. dsh-lost-and-found is a local file memory indexing plugin maintained by wangzhanchao883. It addresses the pain point of local file retrieval: by scanning specified directories to build a SQLite index, it lets users quickly locate files using natural language or keywords, without needing to search through the entire hard drive or rely on cloud storage.

Core Features

The plugin records file information in a local SQLite database to enable efficient local retrieval.

  1. Indexing: Stores scanning information for specified folders (name, type, size, time appeared, location, source) in a local SQLite database.
  2. Incremental Scanning: Each directory maintains its own incremental anchor. The first scan of a new directory backfills historical data, after which it switches to incremental updates.
  3. Multi-dimensional Search: Supports searching by keywords, time range, file type, directory scope, and more. Matching locations are automatically weighted for sorting.
  4. Real-time Fallback: If no results are found in the index, the plugin performs a real-time, read-only scan over the scan directories to try to recover the file.
  5. Image Descriptions: Supports having a model review image content and write back descriptions, so images can later be searched by visual content.
  6. Read-only Mode: The plugin remains read-only at all times and never modifies, moves, or deletes any files.

Installation and Enablement

Installing this plugin requires using the web profile of DSH.

dsh plugin --profile web add dsh-lost-and-found

Typical Usage

You can use natural language directly in the conversation, or call the tools provided by the plugin.

Natural language examples:
* “Where is the quotation I received last month?”
* “Where was the report about XX that I asked you to write stored?”
* “Where did the downloaded installer go? I think it was this week.”

Tool commands:
* file_find: The core file-finding tool. It supports keywords plus time/type/directory/source filters and automatically falls back when the index has no match.
* file_index_scan: Scans once immediately. Supports wait=true to wait until the scan finishes, and full=true to force a full rescan while ignoring anchors.
* file_index_status: Views the scanning overview, including last scan time, record count in the database, scan directories, and more.

Environment Requirements and Configuration

The plugin has specific version requirements for the runtime environment.

  • DeepSeek Harness: Version 0.1.7 ~ 0.2.x, and the web profile must be used.
  • Node.js: Version ^22 or >=24.

Configuration Notes:
* All data is stored only in local SQLite files, and no network connection is required.
* The index only covers configured directories. Files outside those directories can only be found through real-time fallback by name/time.
* New files outside the scanning scope will be reported but are not automatically indexed.

Notes

  • Body and Summary Extraction Not Enabled Yet: Full-text search by document content is not currently supported.
  • Permission Limitations: The plugin runs with the permissions of the current DSH process. Review the source code and license (MIT) before installing.
  • Directory-level Pruning: During scanning, cache directories, VCS directories, and similar locations are automatically skipped to improve efficiency.